The dynamic linear model has the form as

$$ y_t = m(\theta_t, x_t) + \epsilon_t ~~~~~~~~~~~~~~~~~~~~~~~~~~~~ (1)\\ \theta_{t+1} = F \theta_t + R \eta_t,~\eta_t \sim N(\mu_t,\Sigma_t) ~~~~~~(2) $$ where $m(\theta_t, x_t)$ is a mapping function, Equation (1) is known as the observation equation and Equation (2) is known as the hidden state equation.

The question is that if I want to compare the prediction performance between the dynamic linear model and another model, say, linear regression model, is it reasonable to use the cross-validation techniques? Or are there other reasonable techniques?

Out-of-sample method, which uses a block of data at the end of the observed series points, is an option, but I am looking for cross-validation like method.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.